No-Code Agentist

The platform abstraction argument is right, but it's where the story gets interesting for investors, because what Epic is actually doing with Agent Factory is collapsing the distance between "we have an Epic integration" and "we built this ourselves." Health systems sitting on 10+ years of structured EHR data can now prompt their way into agentic workflows without a vendor contract. That changes the procurement conversation entirely, the startup that spent 18 months getting on the Showroom and building a clean FHIR integration suddenly finds itself in a bake-off against a drag-and-drop tool the health system's own IT team built over a quarter. The structured data point you raise is the real lever here. Epic's data model is already normalized across vitals, labs, meds, problem lists, the thing that made third-party AI vendors valuable was partly their ability to make sense of that structure at scale. Agent Factory hands that capability back to the health system directly (which is good for the health system, genuinely, but rough for anyone whose pitch deck assumed Epic would stay in the infrastructure lane). Where I'd push further: the white space that survives this isn't "better Epic integration," it's data Epic never touches. Claims, wearables, genomics, social determinants, pharmacy fill patterns across non-Epic systems. That's the adjacent territory that doesn't get absorbed by a no-code agent builder, because the builder is only as good as the data model underneath it. https://t.co/UCHgUybasS

Fable access just got extended to July 19. The narrative: more time with Anthropic's top model. What the announcement actually introduced that builders should track: 1. Usage limits cut to 50% of your previous plan allocation 2. A new intermediate classifier silently routes "basic code" requests to Opus instead of Fable — with no stated signal for which model actually handled each request The halved limits are annoying. The classifier is the structural change. If Fable is your planning layer in multi-step agent workflows, you're now operating with a black-box router between you and the model. It's no longer "call Fable → get Fable" — it's "call Fable → classifier decides." This matters most when debugging, benchmarking, or when your system prompt is tuned for one model tier and silently runs on another. The extension is real. But the trust model changed. For those building Claude pipelines — are you actively accounting for this routing behavior, or treating it as noise?

HERE’S HOW TO PLAN WITH AN EXPENSIVE MODEL AND EXECUTE WITH A CHEAP ONE one of the simplest multi agent workflows on @october_ai is to get claude fable 5 to plan the task, hand it to grok 4.5, get grok to execute it and hand it back to fable 5 for review no need to hack around claude code settings to get it to use cheaper models for execution! you can just drag a line between two agents on the canvas (bonus: this is multiplayer like figma) #ai